Datasets:
parent string | tokenizer string | vocab_size int64 | dtype string | n_tokens int64 | n_docs int64 | seq_len int64 | n_sequences int64 | v2_prefix_tokens int64 | v2_prefix_seqs int64 | new_perm_seed int64 | source string | sha256 dict | elapsed_s float64 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
data/v2 | NousResearch/Llama-2-7b-hf | 32,000 | uint16 | 8,200,000,000 | 6,954,441 | 2,048 | 4,003,906 | 1,050,000,000 | 512,695 | 1 | HuggingFaceFW/fineweb-edu sample-10BT (stream order) | {
"tokens_u16.bin": "e776fde67843376f4743b8e070a2af9145315b515b262e54003e5e6da38f2afa",
"tokens_u16_doc_offsets.npy": "ccc0c10b020e9f69e63cf20e0a0f79c7358f25b03d148e91dbe33a196d29e92b",
"seq_permutation.npy": "35960673eb0db5bdd2e6f51138d717c97963f0376ca679ae22889d35af416399",
"main_heldout.bin": "23887c3297bff9... | 2,358.5 |
Pre-tokenised language-model training corpus, v3
A fixed, byte-verified training corpus for small-scale language-model scaling experiments: 8,200,000,000 tokens of
FineWeb-Edu sample-10BT in stream order, tokenised with the Llama-2 tokenizer (vocabulary 32,000, uint16, no
BOS/EOS, documents concatenated), plus three held-out evaluation slices. Sequences are 2,048 tokens;
seq_permutation.npy fixes the training order so that a run on the first N sequences is a byte-identical prefix of any
longer run.
| file | content |
|---|---|
tokens_u16.bin |
8.2e9 training tokens (uint16) |
tokens_u16_doc_offsets.npy |
document boundaries |
seq_permutation.npy |
fixed order of the 4,003,906 sequences |
main_heldout.bin |
in-domain held-out slice |
xd_lang.bin (+offsets) |
10M tokens of German web text (FineWeb-2 deu_Latn), cross-domain probe |
xd_code.bin (+offsets) |
10M tokens of Python (the-stack-smol-xs, codeparrot-clean-valid), cross-domain probe |
artifact_meta_v3.json |
manifest with the SHA-256 of every file |
Sources and licences: FineWeb-Edu and FineWeb-2 (ODC-By 1.0); the-stack-smol-xs and codeparrot-clean-valid (permissively licensed source files; see the upstream cards). This repository redistributes token ids only.
- Downloads last month
- 146